Professional Documents
Culture Documents
com
ORIGINAL RESEARCH ARTICLE
The amplitude, frequency and number of oscillating movements have a significant impact on the
possible damage modes of rolling bearings. Changes in amplitude change both the fatigue and wear risk
of the raceways. In order to properly calculate the fatigue lifetime of an oscillating bearing, estimate the
risk of surface-induced damage modes and design test programs for such bearings, it is necessary to
understand the characteristics of the oscillating movements. This paper presents a comprehensive
method for the analysis of time series data including position and loads. The method is applied to the
simulated time-series data of the IWT 7.5-164 reference wind turbine. Wind turbine blades typically
experience wide-band dynamic loads due to stochastic wind conditions. Therefore, it is important to use
time series based cycle counting methods.
1755-0084/ã 2018 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). https://doi.org/10.1016/j.
40 ref.2018.02.004
Renewable Energy Focus Volume 25 June 2018 ORIGINAL RESEARCH ARTICLE
TABLE 1
Main parameters of the IWT 7.5-164 [14].
Property Symbol Value
Wind class IEC A1
Hub height hhub 119.3 m
Rotor diameter (unconed) Drotor 164 m
41
ORIGINAL RESEARCH ARTICLE Renewable Energy Focus Volume 25 June 2018
ORIGINAL RESEARCH ARTICLE
FIGURE 2
IWT7.5-164 wind turbine simulation at 11 m/s mean wind speed.
FIGURE 3
Severe fatigue damage (left) and surface-induced damage (right) on bearing raceways.
amplitudes control the power output of the turbine, whereas the standardized in the ISO 281 [8]. This is the case for long, steady
turbines’ IPC executes oscillations with smaller amplitudes for the rotations with relatively high speed, when elastohydrodynamic
purpose of load reduction. conditions apply. The extent of RCF damage is affected, among
As the simulated time of 600 s is just a short interval of the others things, by the number of times a rolling body passes a
overall lifetime of the turbine, each simulation run of 600 s certain area of the raceway and by the load [17]. The left part of
represents a longer period. This period is obtained by using the Figure 3 shows severe RCF damage on the raceway of a four-point
distribution of the wind speed over 20 years. The IWT7-5164 is contact bearing.
designed for IEC class IA. The IEC 61400-1 [13] contains a descrip- Oscillations with small amplitudes, however, give rise to surface-
tion of the site conditions for the different generic classes and the induced damage. Elastohydrodynamic contact conditions are less
methods for multiplying the simulation data to obtain the loads likely than mixed lubrication regimes. Surface-induced damage
for the full lifetime. might appear in the form of False Brinelling and/or Fretting
Corrosion [18]. The right part of Figure 3 shows surface-induced
Effect of movements on damage modes damage on the raceway of a four-point-contact bearing.
Assuming a bearing is correctly designed and mounted, the two The risk of it occurring, its severity and the speed of damage
main damage modes of pitch bearings are rolling contact fatigue progression depend on several parameters. These are, among
and surface-induced damage modes on the bearing raceways [6]. others, the amplitude and speed of the oscillations, the load
RCF is the dominant mode when a lubricant film separates the and lubricant. Whether an oscillation generates surface-induced
rolling bodies from the raceway and no metal–metal-contact damage also depends on the size of the contact patch between the
occurs. Lundberg and Palmgren first described the fatigue mech- rolling body and raceway as well as the size of the bearing. Thus,
anisms in rolling bearings [1], today fatigue calculation is amplitudes of different bearings should not be compared by means
42
Renewable Energy Focus Volume 25 June 2018 ORIGINAL RESEARCH ARTICLE
of their absolute value in degrees, but rather by the relation Analysis of oscillations
between the rolling distance x and the size of the contact patch As the size of the output files of the simulation makes it inefficient
along this direction 2b (see Figure 4). to use them as a direct input for component design calculation, the
This relation is also called the amplitude ratio. The findings of data have to be processed prior to further design steps. The results
Maruyama and Saitoh for oil-lubricated bearings [19] are used for of this processing are used to calculate the fatigue lifetime of an
the subsequent analysis: In [19], it is shown that an amplitude ratio oscillating bearing, estimate the risk of surface-induced damage
FIGURE 5
Analytical methods: rainflow counting, bin counting, damage equivalent loads.
43
ORIGINAL RESEARCH ARTICLE Renewable Energy Focus Volume 25 June 2018
ORIGINAL RESEARCH ARTICLE
FIGURE 6
Rainflow counting (left) vs. range pair counting (right).
FIGURE 7
Damage equivalent load process.
half cycle 1 is relevant for the structural fatigue damage of materi- amplitudes and cycles (indicated as 1, 2, 3 in the figure). Given
als (see [12] for further details). However, with regard to blade- a certain S/N slope, these values are transferred to an arbitrarily
bearing movements, it is more important to focus on the single chosen amplitude level that is the same for all values (dotted
distances x covered by the roller/ball on the raceway. Every arrows in Figure 7). The cycle values can then be summed (S-sym-
interruption of a movement leads to a breakdown of possible bol in Figure 7). The S/N slope is again used to transfer this
lubricant films and has to be taken into account. As such, the less summed value to a chosen number of cycles and the correspond-
common range-pair counting approach (see the right side of ing amplitude level. None of these operations changes the
Figure 6) is more suitable in this case. Rainflow counting over- summed fatigue damage to the part. As any information about
estimates the number and amplitude of long cycles (see 1 in single amplitude values is erased in the process, DEL results are not
Figure 6) and the number of cycles with relatively small amplitudes suitable for the analysis of blade-bearing movements.
(see 3, 5 and 7 in Figure 6).
In its most basic form, bin counting provides the period of time Program for data analysis
that any given signal spends within a set range of values. Combi- To take account of the range-pair counting method and additional
nations of bins, for example, subsequent counting of second signal bin counting, a MATLAB-program was created for the purpose of
values may be used to obtain more information. As no amplitude blade-bearing movement and load analysis. Figure 8 shows a flow
information is stored, bin counting as a solitary tool is not valuable chart of this program. The raw time series data is used as the input.
in the analysis of blade-bearing movements. Yet, applying bin The blade pitch angle signal is analyzed by a range-pair counting
counting as a supplementary method to the range-pair counting is algorithm which includes an evaluation of the mean value of each
beneficial, as the combination of the two gives information on the half cycle. For each half cycle, additional signals like MxB or MyB
load level for certain oscillating movements, for example. are taken into account via a bin-counting algorithm.
The process of DEL creation is also shown in Figure 7. The results Figure 9 shows and exemplary time series and visualizes the
of the rainflow counting are a number of combinations of results of the analysis. The time series file chosen for this example
44
Renewable Energy Focus Volume 25 June 2018 ORIGINAL RESEARCH ARTICLE
FIGURE 9
Visualization of the analytical results of an exemplary time series.
was one of the six files (seeds) of DLC1.1 with a wind speed of 9 m/s operating time or 60,766 h. This figure corresponds to the IPC
and the wind direction of 0 . The pitch angle signal is shown in controller mode being active at the rated speed and above, as the
blue and contains 13 visually identified half cycles with double Weibull distribution for this wind class implies a probability of
amplitude above 1 . This threshold was set arbitrarily to allow easy 37% for wind speeds of 10 m/s and above. The high number of
visual identification. The amplitude sizes(in the gray boxes of cycles in the smallest amplitude bin (0.03–5 ) is typical for an
Figure 9) are the output of the MATLAB-analysis. The bin counting individual pitch controlled turbine as well.
result of MxB (red signal) is visually interpreted by means of As most pitch activity occurs in the bin with the smallest
colored boxes. The bin counting is executed only when pitch amplitudes (0.03–5 in Table 2), it is further split up into finer
half-cycles are identified. Time periods without colored boxes bins, see Table 3. The amplitude is given both in degrees and
are such periods where no half cycles were identified (either rolling distance covered by the ball on the raceway. Assuming an
standstill or too small an amplitude). amplitude ratio of roughly 1.5 as the lower limit for a lubricant film
formation (cf. ‘Effect of movements on damage modes’ section), a
Results and discussion limit for cycles that impose a risk of surface-induced damage can be
All fatigue load cases according to IEC 61400-1 [13] were analyzed, estimated. This limit is only an orientation value, as further
namely the Design Load Cases 1.2, 2.4, 3.1, 4.1 and 6.4. The research in the field of oscillating movements of grease-lubricated
operating time of these load cases sums up to 175,000 h, that is, contacts is necessary. A value of 0.4 for the double amplitude was
approx. 20 years. Start and stop (DLC 3.1 and 4.1) were assumed to chosen as it was closest to an amplitude ratio of 1.5. The sum of all
occur 1100 times each year based on values of comparable wind cycles with amplitudes equal to or smaller than 0.4 is 1.29 107.
turbines. In several research projects dealing with smaller oscillating bear-
Table 2 shows the results of a range-pair counting of movements ings, surface-induced damage was created with far lower cycle
between 0.03 and 90 width. The lower value is included in the numbers of up to 0.5 106. A list of research projects and the
range, the upper excluded. The overall number of cycles identified kinematic conditions investigated in these can be found in [6].
is 566 107. Taking the mean frequency and the numbers of cycle, Thus, the results confirm that there is a risk of surface-induced
it can be calculated that the pitch system is active for 35% of the damage in the investigated case.
45
ORIGINAL RESEARCH ARTICLE Renewable Energy Focus Volume 25 June 2018
TABLE 2
Cycle count results for 0–90 .
Double amplitude ( ) Number of full cycles (–) Mean frequency (Hz)
0.03–5 4.81E + 07 0.43
5–10 8.11E + 06 0.16
10–15 3.31E + 05 0.11
ORIGINAL RESEARCH ARTICLE
TABLE 3
Detailed cycle count results for 0–5 .
Double amplitude ( ) Double amplitude x (mm) Number of full cycles (–) Mean frequency (Hz)
0.03–0.2 0.6–4 8.66E + 06 0.87
0.2–0.4 4–8 4.25E + 06 0.59
0.4–0.6 8–12.1 3.05E + 06 0.51
0.6–0.8 12.1–16.1 2.32E + 06 0.45
0.8–1.0 16.1–20.1 2.30E + 06 0.40
1.0–1.2 20.1–24.1 2.00E + 06 0.37
1.2 –1.4 24.1–28.2 1.84E + 06 0.33
1.4 –1.6 28.2–32.2 1.63E + 06 0.31
1.6–1.8 32.2–36.2 1.73E + 06 0.30
1.8–2.0 36.2–40.2 1.43E + 06 0.28
2.0–3.0 40.2–60.3 6.37E + 06 0.25
3.0–4.0 60.3–80.5 6.76E + 06 0.21
4.0–5.0 80.5–100.6 5.77E + 06 0.19
TABLE 5
TABLE 4
Bin count of pitch angle mean values. Bin count of resulting bending moment.
Double amplitude ( ) Pitch angle value ( ) Operational time (h) Double Resulting bending Operational
amplitude ( ) moment (kNm) time (h)
0.03–0.40 0–2.5 3450.47
2.5–5 289.81 0.03–0.40 6000–8000 135
5–10 658.67 8000–10,000 205
10–15 553.69 10,000–12,000 286
15–20 198.00 12,000–14000 401
20–25 45.27 14,000–16,000 601
16,000–18,000 928
0.40–1.00 0–2.5 2979.88 18,000–20,000 1116
2.5–5 534.41 20,000–22,000 924
5–10 964.93 22,000–24,000 452
10–15 720.15 24,000–26,000 106
15–20 310.73
20–25 64.01
1.00–5.00 0–2.5 6208.65
2.5–5 3950.09 Table 4 shows the results of this bin count. For double amplitudes
5–10 9480.14 smaller than 0.4 , 66% of the oscillations have a mean value
10–15 9417.40 between 0 and 2.5 . This range contains the largest single operat-
15–20 4595.34
ing time bin in the evaluated values. Sixty-six percent equals
20–25 1290.96
0.86 107 cycles, a number still high enough to impose a risk
of surface-induced damage.
A high number of oscillations with small amplitudes increases The analysis of pitch angle amplitudes indicates the risk of
the risk of surface-induced damage, especially if the mean values surface induced damage for this blade bearing. Whether such
do not change over time. A changing mean value would result in a damage occurs also depends on various other factors, one of them
change of the raceway areas that are in contact with the rolling being the applied load. To gain an understanding of the load
bodies. These areas share the oscillations/loads, thus reducing the conditions, a bin count of the resulting bending moment was
probability that one individual area will fail. To better understand carried out. Table 5 contains the results of this bin counting.
the mean value behavior of the pitch angle, a bin count of the The analytical result is a range pair count including oscillation
pitch angle was performed in addition to the range pair count. frequencies and accompanying bin count of pitch angle and
46
Renewable Energy Focus Volume 25 June 2018 ORIGINAL RESEARCH ARTICLE
bending moments. This result can be used as input for a rolling this work will be a movement pattern analysis of data and incor-
contact fatigue calculation (see [2]) or to design a test program for poration of long term wind measurements. Advanced pattern
either a test of surface-induced damage modes or rolling contact recognition, in particular, will allow for a better understanding
fatigue. For fatigue calculations, it is recommended to use a higher of critical conditions. The work will also include bearing tests
number of load bins, as otherwise the design of the bearing would under different oscillating movements to identify critical ampli-
be too conservative [21]. The MATLAB-program we use for this tude ratios under these conditions.
47